Localized Modeling for Airline Price Prediction Using K-Means and Decision Tree Ensemble

Abstract

Both travelers and airline companies rely on accurate prediction of flight prices, nevertheless it is difficult to train machine learning models using large-scale, current flight datasets due to computational inefficiency and risk of overfitting. This paper introduces a novel two-pronged approach that combines K-means clustering with decision trees for effective localized flight price forecasting. First, K-means clustering is employed to segment the flight data on the basis of shared characteristics into meaningful clusters thereby reducing data dimensionality and speeding up model training. Then individual Decision Tree models are built for each cluster separately. Finally, this technique emphasizes on closely related data attributes which may enhance predictive accuracy for given routes or types of flights. The proposed method solves the problem of calculation load when working with large datasets without sacrificing any details necessary for catching peculiarities in pricing in various categories of flights.